P42/mnm

Vollständiger Abstract

Worum geht es in dieser Arbeit?

BACKGROUND: This study aimed to develop a single-photon emission computed tomography/computed tomography (SPECT/CT)-based radiomics model using bone metastases to predict epidermal growth factor receptor (EGFR) mutation status and the efficacy of targeted therapy in patients with non-small-cell lung cancer. METHODS: This retrospective study included 240 patients with non-small-cell lung cancer. Among them, 205 were analyzed for the EGFR-mutation prediction model, and 210 for the targeted therapy efficacy model. Baseline and post-treatment whole-body SPECT/CT bone scans with maximum standardized uptake value (SUVmax) were acquired. Radiomic features from bone-metastasis SPECT/CT images were selected by minimum redundancy maximum relevance and least absolute shrinkage and selection operator regression. Seven machine learning algorithms were assembled to construct predictive models, whose performance was assessed via receiver operating characteristic curves, calibration curves, decision-curve analysis, and SHapley Additive exPlanations for model interpretability. RESULTS: Among the seven machine learning models, the eXtreme Gradient Boosting model was the optimal. For the clinical-radiomics-SUVmax EGFR-mutation model, predictive factors included BMI, nonsmoking status, aspartate aminotransferase, leukocyte counts, erythrocyte counts, and SUVmax; the area under the receiver operating characteristic curve (AUC) values were 0.867 [95% confidence interval (CI): 0.823-0.911, training] and 0.834 (95% CI: 0.757-0.911, validation). For the radiomics-ΔSUVmax efficacy model (ΔSUVmax > -0.18), AUCs reached 0.869 (95% CI: 0.828-0.910, training) and 0.846 (95% CI: 0.769-0.923, validation) (all P < 0.05). CONCLUSION: SPECT/CT-derived bone metastasis radiomics offers complementary information for predicting EGFR mutation and targeted therapy efficacy. Models integrating clinical and metastatic‑lesion features may facilitate individualized treatment decision‑making.

Abstract: PubMed · Datensatz